Dynamic Time Warping for Pattern Detection in Time Series Flight Data

Rotorcraft with HUMS systems typically generate a large amount of time series data such as ambient air temperatures or various engine temperatures. For a fleet over time, this data set can become quite large and can't all be consumed by the human eye in the form of strip charts in a timely manner. When an anomaly occurs, there is a likely chance that a subject matter expert (SME) will want to find out if that particular anomaly has ever happened before and when it happened. The SME will want to find anything in the history of the fleet that looks like the anomaly that they just saw. This paper presents a method called dynamic time warping (DTW) for doing just that. A SME can input a query array and the DTW algorithm will find all occurrences that are similar to the one the SME input. The result is a drastically reduced set of data for the SME to physically look at. DTW will be introduced and explained and an example use of DTW on time series flight data will be presented along with tips and tricks for expanded use of this type of algorithm.